IF YOU WOULD LIKE TO GET AN ACCOUNT, please write an
email to Administrator. User accounts are meant only to access repo
and report issues and/or generate pull requests.
This is a purpose-specific Git hosting for
BaseALT
projects. Thank you for your understanding!
Только зарегистрированные пользователи имеют доступ к сервису!
Для получения аккаунта, обратитесь к администратору.
Only notify the first 5 pending lock waiters. This avoids a thundering herd
problem that is really nasty in a cluster. It also makes acquiring a lock a bit
more FIFO, lock waiters are added to the end of the array.
Only check the existence of the lock owner in g_lock_parse, check the rest of
the records only when we got the lock successfully. This reduces the load on
process_exists which can involve a network roundtrip in the clustered case.
This made smbd crash in g_lock_lock() when trying to start a
transaction on a db with an already started transaction,
e.g. in a tcon_and_X where the share_info.tdb was not yet
initialized but share_info.tdb was already locked by another
process or writing acces to the winreg rpc pipe where the
registry tdb was already locked by another process.
What we really _want_ to do here by design is to react to
MSG_DBWRAP_G_LOCK_RETRY messages that are either sent
by a client doing g_lock_unlock or by ourselves when
we receive a CTDB_SRVID_SAMBA_NOTIFY or
CTDB_SRVID_RECONFIGURE message from ctdbd, i.e. when
either a client holding a lock or a complete node
has died.
Doing this properly involves calling tevent_loop_once(),
but doing this here with the main ctdbd messaging context
creates a nested event loop when g_lock_lock() is called
from the main event loop.
So as a quick fix, we act a little corasely here: we do
a select on the ctdb connection fd and when it is readable
or we get EINTR, then we retry without actually parsing
any ctdb packages or dispatching messages. This means that
we retry more often than necessary and intended by design,
but this does not harm and it is unobtrusive. When we have
finished, the main loop will pick up all the messages and
ctdb packets. The only extra twist is that we cannot use
timed events here but have to handcode a timeout for select.
Michael
This skips update of the __db_sequence_number__ record when nothing else has
been written. There are transactions that are just openend and then nothing
is written until transaction_commit is called. This is for instance the case
with registry initialization routines: They start a transaction and only
write somthing when the registry has not been initialized yet.
So this change will skip many db_seqnum bumps and TRANS3_COMMIT roundtrips.
Michael
I carefully prepared the return value only to "return 0;" at the bottom. :-(
This may well have hit us for instance in the nested cancel case
and produced random errors.
Michael
The logic bug was that if a record was found in the marshall buffer,
then always the ctdb header of tha last record in the marshall buffer
was returned, and not the ctdb header of the last occurrence of the
requested record.
This is fixed by introducing an additional temporary variable.
Michael
For persistent databases, 64bit integer is kept in a special record
__db_sequence_number__. This record is incremented with each completed
transaction.
The retry mechanism for failing TRANS3_COMMIT controls inside the
db_ctdb_transaction_commit() function now relies one a modified
behaviour of ctdbd's treatment of persistent databases in recoveries.
Recently, a special treatment for persistent databases had been
introduced in ctdb (1.0.108) to work around the problems with the
orinal design of persistent transactions.
Now with the rewrite we need to revert to the old behaviour that
ctdb always takes the newest copies of all records.
This change also paves the way for a next step, which will make
recovery use the db seqnum to tell which node has the newest copy
of a persistent db and use that node's copy. This will greatly
reduce the amount of data transferred with each recovery.
Michael
This simplifies the transaction code a lot:
* transaction_start essentially consists of acquiring a global lock.
* No write operations at all are performed on the local database
until the transaction is committed: Every store operation is just
going into the marshall buffer.
* The commit operation calls a new simplified TRANS3_COMMIT control
in ctdb which rolls out thae changes to all nodes including the
node that is performing the transaction.
Michael
This is the basis to implement global locks in ctdb without depending on a
shared file system. The initial goal is to make ctdb persistent transactions
deterministic without too many timeouts.
This changes the meaning of the ->prev pointer in our doubly linked
lists to point at the end of the list from the front of the list. That
allows us to implement DLIST_ADD_END() and related functions in O(1)
time, which can be a huge saving in many places in Samba.
This also means that the 'type' argument to various DLIST_*() macros
is no longer needed, but I have left it in for now to keep the
patchset small, which will make it easier to revert if any problems
are found. In the future we should remove the 'type' arguments.
(jra. Move the one use of DLIST_TAIL over to the new macros).
tevent ensures that a timed event is only called once. The old events
code relied on the called handler removing the event itself. If the
handler removed the event after calling a function which invoked the
event loop then the timed event could loop forever.
This change makes the two timed event systems more compatible, by
allowing the handler to free the te if it wants to, but ensuring it is
off the linked list of events before the handler is called, and
ensuring it is freed even if the handler doesn't free it.
Only works on Linux kernels 2.6.26 and above. Grants CAP_KILL capability
to allow Linux threads under different euids to send signals to each other.
Jeremy.
